Forecasting COVID-19 Outbreak in India Using Time Series Dataset: An Ensemble of ARIMA, Abbasov-Mamedova, and Multilayer Perceptron Models

Chakraborty, A.; Mitra, S.; Das, D.; De, D.; Pal, A. J.

Recently the COVID-19 pandemic outbreak is inflicting devastation on human civilization. This infectious virus spreads like wildfire, already affected millions worldwide, and the numbers are still increasing. This situation warrants a comprehensive strategy backed by futuristic estimations to counter COVID-19 adversities. Like any other country globally, India is also encountering an uphill task to fight against this unfortunate pandemic, with six million-plus COVID-19 cumulative infected cases by the first week of October 2020. This publication elucidates the use of four state-of-art models, namely the Abbasov - Mamedova (AM) Fuzzy, proposed Multilayer Perceptron (MLP), Auto-ARIMA, and Auto-MLP, to forecast the number of cumulative infected COVID-19 cases in India. These models exhibited high forecast accuracy for 30 days ahead scenario with MAPE ranges from 0.44 to 1.83% in the test condition, whereas a MAPE range of 1.09 to 2.39% in real-time. We estimated the COVID-19 cases fortnightly and observed that the proposed MLP exhibited the flattening of the COVID-19 curve, whereas other models exhibited a rising trend. Though our proposed MLP outperformed other models, we employed all four methods and estimated a range between 8.53 to 13.77 million COVID-19 positives by 4th January 2021 in India. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.